Abstract
Smallholder farmers in Sub-Saharan Africa face severe and persistent information barriers. A new wave of AI-assisted digital information technologies has the potential to overcome these barriers at scale, delivering personalized, hyperlocal agronomic advice at low cost. Yet little is known about the impacts of these technologies and how they diffuse. This pilot study investigates these questions in the context of a novel AI-assisted agricultural advisory tool called Virtual Agronomist. Farmers can use this technology to generate tailored nutrient management plans based on high-resolution soil maps, diagnose plant health and pest problems, and access weather advisories. We investigate the impacts of this tool on farmer practices and agricultural outcomes. We also use this context to study broader questions about how information technologies diffuse. We explore three potential mechanisms. First, because many such technologies deliver information as their primary output, peer adoption may generate information spillovers that permanently substitute for own adoption. Second, in contrast to canonical models where heterogeneity slows diffusion, agricultural heterogeneity may reduce the value of free-riding and accelerate adoption. Finally, information technologies constitute a new information source, changing incentives to form social network connections and potentially amplifying or dampening diffusion. We conduct a randomized experiment across 30 villages in Butaleja District, Eastern Uganda to explore these mechanisms. In 10 villages, farmers adopt the tool directly on their own phones. In a second set of 10 villages, usage is mediated by lead farmers who operate the tool on their behalf. The final 10 villages serve as a control, allowing us to compare adoption rates, feature usage, and agricultural outcomes across dissemination strategies.